Getting Started with AI Coding Tools

Getting Started With AI Coding Tools

CBorg supports several AI-powered coding environments. This page helps you choose the right tool and walks you through the setup steps.

Is my tool supported?

Start with the tool you want to use and read across to the access route you plan to bill against. The table below covers every coding tool documented on this site, plus the widely-requested tools that do not work with CBorg, so that a missing entry is not mistaken for an untested one.

Access routes

An access route is how you reach a model and who gets billed for it. The same model can be reachable by more than one route, at different prices and with different responsibilities.

CBorg API

Managed IT Division proxy. One endpoint for on-prem and commercial models.

Models

OpenAI, Anthropic Claude, Google Gemini, xAI, plus free on-prem models

Billed to

CBorg key budget

Free allowance

$50/month per user, plus zero-cost on-prem models

Endpoint

https://api.cborg.lbl.gov

Before you start

CBorg API key

Standard retail per-token pricing for commercial models. On-prem models (lbl/cborg-coder, lbl/cborg-coder-fast, gpt-oss-120b and others) have no per-token cost.

More detail: /api_cborgvscloud/

GCP Cloud Project

Self-managed cloud project attached to your PID, using Vertex AI directly.

Models

Google Gemini, Anthropic Claude, AWS Nova, Cohere and others. No OpenAI models.

Billed to

Your GCP project

Free allowance

$25/month per project

Endpoint

Vertex AI (per-project endpoints)

Before you start

Cloud Project request (email Science IT)
Vertex AI application default credentials or a service account

Per-token inference may be around 10% lower than CBorg retail pricing. You are responsible for budget alerts, cost controls, service accounts and model enablement. On-prem models are not available on this route.

More detail: /api_cborgvscloud/

What the statuses mean

Supported: Setup is documented on this site and the tool's normal agent workflow works.
Limited: Connects and responds, but important modes are missing or degraded. Read the note.
Deprecated: Previously worked. No longer maintained or superseded by another tool.
Not available: Known not to work on this route, or the route cannot serve the required models.
Not documented: No setup guide and no test result. It may work; nobody has confirmed it.

Tool support

Coding tool support by access route. Each cell gives the support status, what you configure, and any caveats.
Tool	CBorg API	GCP Cloud Project	Other routes
Zoo Code VS Code extension	Supported LiteLLM provider Select the LiteLLM API provider and set the base URL to the CBorg endpoint. Source	Supported Vertex AI credentials A Vertex AI provider is added automatically when credentials are detected at ~/.zoo/application_default_credentials.json or via GOOGLE_APPLICATION_CREDENTIALS. Source	American Science Cloud models via AMSC_I2_API_KEY.
Cline VS Code extension	Supported LiteLLM provider Use the LiteLLM provider, not “OpenAI Compatible” — the latter results in a reduced feature set. Source	Supported Select GCP Vertext AI as provider.	—
Claude Code CLI and IDE extension	Supported Anthropic base URL Set ANTHROPIC_BASE_URL and ANTHROPIC_AUTH_TOKEN. Non-Claude CBorg models can be mapped onto the Haiku/Sonnet/Opus slots with the ANTHROPIC_DEFAULT_*_MODEL variables to control cost. Source	Supported Claude Code can be configured to use GCP by setting certain environment variables. Source	Anthropic subscription or API account, billed by Anthropic.
Codex CLI CLI	Supported OpenAI-compatible base URL Add a [model_providers.cborg] block in ~/.codex/config.toml pointing at the CBorg /v1 endpoint, then reference it from a profile. Source	Not available Cloud Projects do not provide OpenAI models. Source	OpenAI API account or a ChatGPT plan.
Codex App Desktop app	Not available Does not work with the CBorg API. Source	Not available Cloud Projects do not provide OpenAI models. Source	ChatGPT Enterprise.
Opencode CLI	Supported Patched build Requires the Science IT build with the LiteLLM provider patch until upstream support is merged. Prebuilt binaries and a build script are provided. Gives accurate cost tracking once installed. Source	Not documented Not tested against an LBL Cloud Project.	—
Aider CLI	Supported OpenAI-compatible base URL Set OPENAI_API_BASE to the CBorg endpoint and OPENAI_API_KEY to your CBorg key. Source	Not documented Not tested against an LBL Cloud Project.	—
VS Code GitHub Copilot Chat with BYOK	Limited OpenAI-compatible base URL Add a model through the “OpenAI Compatible” provider. Support is partial and setup documentation is still in progress. Source	Not documented Not tested against an LBL Cloud Project.	GitHub Copilot subscription for the bundled models.
Cursor Standalone editor	Not available OpenAI-compatible base URL Since Cursor calls CBorg from its AWS servers, it gets blocked by the LBL firewall. There is currenly no workaround for this. Source	Supported Vertex AI credentials Two separate routes. Cursor can be configured to use GCP models directly through its own model settings. Independently of that, the Claude Code plugin can be installed in Cursor and pointed at Claude models served from Vertex AI – the linked source documents that second route only. Source	Cursor subscription, which is not available through CBorg.
Gemini CLI CLI	Deprecated Gemini base URL Previously configured with GOOGLE_GEMINI_BASE_URL and GEMINI_API_KEY. Instructions are retained for reference only and may not work. Source	Not documented Not tested against an LBL Cloud Project.	—
Antigravity CLI, Agent Standalone and IDE	Not available Does not work with the CBorg API. Source	Supported Antigravity (the agent standalone product, not Antigravity IDE) can be used with a GCP Cloud Project.	—
Continue VS Code extension	Deprecated OpenAI-compatible base URL Works with CBorg, but documentation and setup instructions are no longer maintained due to low usage. Source	Not documented Not tested against an LBL Cloud Project.	—
PyCharm, IntelliJ, and other JetBrains IDEs IDE	Supported As of version 2026.2.x, CBorg models are supported in the JetBrains IDEs.	Supported Works with Gemini Enterprise API keys.	—
LBL Omni-Engineer CLI	Supported CBORG_API_KEY environment variable Define CBORG_API_KEY before launching. Works with on-prem CBorg models as well as commercial ones. Source	Not documented Not tested against an LBL Cloud Project.	—
Jupyter AI JupyterLab extension	Supported CBorg provider plugin Install the cborg-jupyter-ai model provider module. Source	Not documented Not tested against an LBL Cloud Project.	—
RStudio gptstudio addin	Limited OpenAI-compatible base URL Set OPENAI_API_URL to the CBorg /v1 endpoint. Requires the development version of gptstudio; the released version did not support OpenAI-compatible servers. Source	Not documented Not tested against an LBL Cloud Project.	—
GitHub Copilot IDE extension	Not available Not supported through CBorg. See VS Code above for the BYOK route that does work. Source	Not available Copilot does not accept external model credentials in its standard configuration.	GitHub Copilot subscription.

Zoo Code

VS Code extension

Interactive GUI agent with visual diff approval. Formerly Roo Code.

CBorg API: Supported LiteLLM provider
Select the LiteLLM API provider and set the base URL to the CBorg endpoint.
Source
GCP Cloud Project: Supported Vertex AI credentials
A Vertex AI provider is added automatically when credentials are detected at ~/.zoo/application_default_credentials.json or via GOOGLE_APPLICATION_CREDENTIALS.
Source
Other routes: American Science Cloud models via AMSC_I2_API_KEY.

Cline

VS Code extension

Interactive VS Code agent. Sibling project to Zoo Code.

CBorg API: Supported LiteLLM provider
Use the LiteLLM provider, not “OpenAI Compatible” — the latter results in a reduced feature set.
Source
GCP Cloud Project: Supported
Select GCP Vertext AI as provider.

Claude Code

CLI and IDE extension

Anthropic's agentic CLI. Strong on large codebases and multi-file work, and among the most expensive to operate.

CBorg API: Supported Anthropic base URL
Set ANTHROPIC_BASE_URL and ANTHROPIC_AUTH_TOKEN. Non-Claude CBorg models can be mapped onto the Haiku/Sonnet/Opus slots with the ANTHROPIC_DEFAULT_*_MODEL variables to control cost.
Source
GCP Cloud Project: Supported
Claude Code can be configured to use GCP by setting certain environment variables.
Source
Other routes: Anthropic subscription or API account, billed by Anthropic.

Codex CLI

CLI

OpenAI's coding agent. Best native fit for GPT models. Profiles switch models at runtime.

CBorg API: Supported OpenAI-compatible base URL
Add a [model_providers.cborg] block in ~/.codex/config.toml pointing at the CBorg /v1 endpoint, then reference it from a profile.
Source
GCP Cloud Project: Not available
Cloud Projects do not provide OpenAI models.
Source
Other routes: OpenAI API account or a ChatGPT plan.

Codex App

Desktop app

OpenAI's desktop surface for computer-use and conversational workflows.

CBorg API: Not available
Does not work with the CBorg API.
Source
GCP Cloud Project: Not available
Cloud Projects do not provide OpenAI models.
Source
Other routes: ChatGPT Enterprise.

Opencode

CLI

Open-source CLI agent. Fully open alternative to Claude Code.

CBorg API: Supported Patched build
Requires the Science IT build with the LiteLLM provider patch until upstream support is merged. Prebuilt binaries and a build script are provided. Gives accurate cost tracking once installed.
Source
GCP Cloud Project: Not documented
Not tested against an LBL Cloud Project.

Aider

CLI

CLI pair-programming agent driven from the terminal.

CBorg API: Supported OpenAI-compatible base URL
Set OPENAI_API_BASE to the CBorg endpoint and OPENAI_API_KEY to your CBorg key.
Source
GCP Cloud Project: Not documented
Not tested against an LBL Cloud Project.

VS Code

GitHub Copilot Chat with BYOK

Copilot Chat's bring-your-own-key (BYOK) mode can be pointed at CBorg. BYOK means you supply the model credentials rather than using the vendor's bundled inference.

CBorg API: Limited OpenAI-compatible base URL
Add a model through the “OpenAI Compatible” provider. Support is partial and setup documentation is still in progress.
Source
GCP Cloud Project: Not documented
Not tested against an LBL Cloud Project.
Other routes: GitHub Copilot subscription for the bundled models.

Cursor

Standalone editor

AI-first VS Code fork with its own model integration.

CBorg API: Not available OpenAI-compatible base URL
Since Cursor calls CBorg from its AWS servers, it gets blocked by the LBL firewall. There is currenly no workaround for this.
Source
GCP Cloud Project: Supported Vertex AI credentials
Two separate routes. Cursor can be configured to use GCP models directly through its own model settings. Independently of that, the Claude Code plugin can be installed in Cursor and pointed at Claude models served from Vertex AI – the linked source documents that second route only.
Source
Other routes: Cursor subscription, which is not available through CBorg.

Gemini CLI

CLI

Google's terminal agent. Superseded by Antigravity CLI.

CBorg API: Deprecated Gemini base URL
Previously configured with GOOGLE_GEMINI_BASE_URL and GEMINI_API_KEY. Instructions are retained for reference only and may not work.
Source
GCP Cloud Project: Not documented
Not tested against an LBL Cloud Project.

Antigravity

CLI, Agent Standalone and IDE

Google's coding agent, successor to Gemini CLI.

CBorg API: Not available
Does not work with the CBorg API.
Source
GCP Cloud Project: Supported
Antigravity (the agent standalone product, not Antigravity IDE) can be used with a GCP Cloud Project.

Continue

VS Code extension

VS Code assistant. Documentation retained for historical context.

CBorg API: Deprecated OpenAI-compatible base URL
Works with CBorg, but documentation and setup instructions are no longer maintained due to low usage.
Source
GCP Cloud Project: Not documented
Not tested against an LBL Cloud Project.

PyCharm, IntelliJ, and other JetBrains IDEs

IDE

The IDE's AI Chat feature can be configured for CBorg and other model providers.

CBorg API: Supported
As of version 2026.2.x, CBorg models are supported in the JetBrains IDEs.
GCP Cloud Project: Supported
Works with Gemini Enterprise API keys.

LBL Omni-Engineer

CLI

Science IT's CLI assistant. Edits files in place using simple commands.

CBorg API: Supported CBORG_API_KEY environment variable
Define CBORG_API_KEY before launching. Works with on-prem CBorg models as well as commercial ones.
Source
GCP Cloud Project: Not documented
Not tested against an LBL Cloud Project.

Jupyter AI

JupyterLab extension

Jupyternaut assistant inside your own JupyterLab instance.

CBorg API: Supported CBorg provider plugin
Install the cborg-jupyter-ai model provider module.
Source
GCP Cloud Project: Not documented
Not tested against an LBL Cloud Project.

RStudio

gptstudio addin

R users can drive CBorg models from RStudio.

CBorg API: Limited OpenAI-compatible base URL
Set OPENAI_API_URL to the CBorg /v1 endpoint. Requires the development version of gptstudio; the released version did not support OpenAI-compatible servers.
Source
GCP Cloud Project: Not documented
Not tested against an LBL Cloud Project.

GitHub Copilot

IDE extension

GitHub's bundled completion and chat service.

CBorg API: Not available
Not supported through CBorg. See VS Code above for the BYOK route that does work.
Source
GCP Cloud Project: Not available
Copilot does not accept external model credentials in its standard configuration.
Other routes: GitHub Copilot subscription.

Connection mechanisms

The mechanism is what you actually configure in the tool. Where a tool offers more than one, the mechanism named above is the recommended choice.

LiteLLM provider: Native LiteLLM provider pointed at the CBorg endpoint. Preferred where available: reports accurate per-request cost and supports prompt caching for Anthropic models. Choosing "OpenAI Compatible" instead gives a reduced feature set.
OpenAI-compatible base URL: Override the OpenAI base URL and API key. Widely supported, but cost reporting and prompt caching are usually less accurate than the LiteLLM provider.
Anthropic base URL: Set ANTHROPIC_BASE_URL and ANTHROPIC_AUTH_TOKEN.
Gemini base URL: Set GOOGLE_GEMINI_BASE_URL and GEMINI_API_KEY.
Vertex AI credentials: Application default credentials or a service account key for your cloud project. Billed to the project, not to a CBorg key.
CBORG_API_KEY environment variable: The tool reads the key from the environment; no endpoint configuration needed.
CBorg provider plugin: A CBorg-specific plugin or model provider supplies the connection.
Patched build: Requires a custom build until upstream support is merged. Prebuilt binaries are provided by Science IT.
Vendor subscription: Access comes bundled with the vendor's own paid service. Not routed through CBorg and not billed to a CBorg key or a cloud project.

Supported Coding Tools

ZooCode – GUI-style / Interactive

Zoo Code (formerly RooCode) is a VS Code extension that provides a fully interactive, GUI-driven coding assistant. It integrates directly into the VS Code editor with a visual diff approval workflow.

Best for: Users who prefer a point-and-click interface and an intuitive “close to the code” experience.

Note: Configuration and optimizing your workflow can be complicated. The extension is under active development.

Claude Code – Deep Development

Claude Code is Anthropic’s agentic coding tool. It is widely used for complex, multi-file development tasks and has strong support for large codebases. Note that Claude models are among the most expensive to operate – average usage is around $6/day, and costs can be higher with suboptimal configuration or heavy usage.

Best for: Power users comfortable with the terminal, deep refactoring and multi-step development tasks.

Note: The Claude Code VSCode Extension also enables in-editor diff viewing and a chat-style sidebar with better readability than the terminal interface.

Note: Review the cost optimization tips before starting to avoid unexpected budget consumption.

OpenAI Codex – GPT Models, Computer Use, Vibe Coding

Codex CLI is OpenAI’s coding agent with the best native integration with GPT models. It also supports the Codex App for computer-use and “vibe coding” workflows. Profiles can be switched at runtime to use different models (GPT, Gemini, etc.).

Best for: Users who prefer GPT models, computer-use workflows, or the Codex App experience.

Opencode – Open Source Power-User’s Alternative (Experimental)

Opencode is a free, open-source CLI coding agent. CBorg support is experimental and requires a custom build until upstream LiteLLM provider support is merged. Recommended for advanced users comfortable building from source.

Best for: Advanced users who want a fully open-source alternative to Claude Code.

VS Code Insiders – Copilot Chat with BYOK

VS Code Insiders supports GitHub Copilot Chat with Bring Your Own Key (BYOK), allowing you to route Copilot requests through the CBorg API. Limited support for the CBorg API is possible using the OpenAI-Compatible provider mode. Setup documentation is in progress.

Budget-Constrained or On-Prem Only?

If you need to keep all work on-premises (no cloud providers) or want to minimize API costs, use the CBorg on-prem models:

lbl/cborg-coder – Full-size on-prem coding model
lbl/cborg-coder-fast – Smaller, faster on-prem coding model

These models run entirely within Lawrence Berkeley National Laboratory infrastructure and do not transmit your data to any 3rd party cloud provider. Any tool marked Supported for the CBorg API in the table above can select them; the on-prem models are ordinary CBorg model names, so no separate configuration is needed. Note that they are not available through a GCP Cloud Project. See the CBorg Free Models page for details.

Step-by-Step Setup Guide

1. Get an API Key

Request a CBorg API key from the API Key Request page. You will need to log in with your Lab-connected Google identity and agree to the terms of service.

2. Request an Increased Budget (Optional)

Keys are issued with a default monthly budget. If your project requires more capacity, you can request a budget increase.

3. Set Up Your Coding Environment

Consult the documentation for your chosen tool:

Tool	Documentation
ZooCode	ZooCode Setup Guide
Claude Code	Claude Code Setup Guide
OpenAI Codex	Codex Setup Guide
Opencode	Opencode Setup Guide
VS Code Insiders	VS Code Insiders Setup

4. Enable CBorg MCP Servers (Optional)

CBorg provides hosted MCP Servers that extend your coding agent with additional capabilities:

Context7 – Up-to-date library documentation and code examples, fetched at query time
Valency – Literature research assistant using preprint servers

MCP servers are supported in ZooCode, Claude Code, Codex, and Opencode. See the MCP Servers page for configuration instructions.

5. Get Help

CBorg Users Chatroom – Join the CBorg Users Google Chat room to ask questions and share tips with other users.
Office Hours – Science IT hosts regular office hours. Contact scienceit@lbl.gov for the schedule.

Other Tools (Experimental)

The following tools have experimental or community-supported CBorg integration. They may require additional testing and configuration. Feel free to reach out to Science IT support if you get stuck.

Windows Subsystem for Linux (WSL) – See the WSL2 guide for running Linux-based coding tools on Windows.
RStudio – R users can connect RStudio to the CBorg API. See the RStudio guide for setup instructions.

For a full list of tools with varying levels of support, see the Legacy / Unsupported Tools page.